Ravi, I spent some time to test your most recent version of this patch. Using the very simple test environment that I described before consisting of just two machines talking to each other with a single level of tagging I encountered several issues. These included not only the hardware offloading issues that I mentioned previously but also an issue with the bottom of stack bit that prevents non-IP traffic from passing. Encountering these types of issues with the first and most basic experiment that I ran gives me significant concern about the level of testing that this patch has seen. I understand that you might have expectations about common use cases but the fact is that Open vSwitch and OpenFlow are designed to encourage new usage models and so it is important that any code we add is robust in all situations. Generally, open source projects would not add a work-in-progress patch and would expect that any patches sent for review consist of what the author considers a final revision. Unfortunately, this means without significant improvements we're not going to able to accept this patch or continue reviewing it (which is quite time consuming).
<rk> Could you please elaborate on testing? Assuming IP==IPV4, non-IP could be IPV6 or any other layer-3. For IPV6: There was a brief discussion in the mailing list regarding this topic. Though I haven't seen any practical usage of IPV6 over MPLS directly, code takes care of handling IPV6 traffic. For any other layer-3: Spec is not clear on how to handle non-IP traffic and I guess it would be at the discretion of the vendor. Will check what the spec says on this. Your testing will probably turn out be VPLS? I believe you have configured flows such that it matches src/dst mac address and actions include mpls related ones. Some of the mpls actions will not apply for this case, however, basic push/pop should work. Code is written such that it should be able to handle this case or easily extensible. I haven't tested this case because I have focused mainly on IPV4->MPLS and to some extent IPV6->MPLS and integration with VLAN and VLAN-qinq. When you say non-IP, what type of traffic is it and how are you generating that traffic? Is it possible to get a remote access to you testbed to understand better? Or if you can give additional information on your testing it would be helpful and I can fix it. My focus is mainly on IP(4/6) over MPLS and VLAN integration and that's what I have tested. Regarding the patch, I am confused myself which patch has what changes since I am focusing on mpls/vlan-qinq/automation in parallel and I am frequently rebasing the code. So it's possible that some of my changes are missing the transition. Once you provide the details of your testing I will take a look at it. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
